Intelligent prediction of eco-concrete compressive strength based on SHAP-interpretable BO-XGBoost and NSGA-II multi-objective optimization
Year: 2026
Type: article
Abstract: Accurate prediction and optimization of eco-concrete compressive strength are important for balancing mechanical performance and material utilization. Based on 400 experimental records, this study developed an interpretable prediction-optimization framework integrating Optuna-TPE-optimized machine-learning models, SHAP, and NSGA-II. Seven mix-design and curing variables were used as inputs, and BO-XGBoost, BO-RF, and BO-SVR were developed for compressive strength prediction. On the independent test set, BO-XGBoost, BO-RF, and BO-SVR achieved R 2 values of 0.958, 0.957, and 0.950, RMSE values of 0.466, 0.475, and 0.509 MPa, and MAE values of 0.376, 0.382, and 0.392 MPa, respectively. BO-XGBoost therefore showed the best overall test-set performance among the three models under the present evaluation procedure. SHAP analysis identified curing age and porosity as the dominant contributors to the model predictions, followed by cement dosage and water-to-binder ratio, and further revealed nonlinear model-attribution patterns across their value ranges. BO-XGBoost was subsequently embedded in NSGA-II to maximize predicted 28-day compressive strength while minimizing cement dosage and water-to-binder ratio. The resulting Pareto front demonstrated clear trade-offs among these objectives. Under the baseline engineering preference, the recommended compromise solution had a water-to-binder ratio of 0.250 and cement dosage of 351.47 kg/m 3 , with a predicted compressive strength of 10.87 MPa. The proposed framework provides an interpretable decision-support approach for eco-concrete mix design.
Source: Scientific Reports
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7223227061